Artificial Intelligence–based investigation of filler selection strategies.

OBJECTIVE: Lineup construction relies on matching fillers to the suspect's appearance or to the eyewitness's description of the perpetrator (match to description, MTD). Recent work shows that low (vs. high)-similarity fillers beyond MTD improve discriminability. We tested lineups constructed with differing suspect-filler (SF) similarity levels beyond MTD (low, moderate, high) against MTD only, varying exposure duration, culprit race, and innocent suspect-culprit resemblance (guilty-innocent [GI] similarity). HYPOTHESES: We expected the highest discriminability for low-SF similarity lineups, moderate next, MTD-only lower, and high lowest-alongside higher discriminability for long (cf. short) exposures, same-race (cf. cross-race) culprits, and low GI similarity (cf. high). We expected high-SF similarity to better protect high GI similarity innocents and low F-similarity to aid culprit identifications regardless of GI similarity. We also anticipated stronger benefits of low-SF similarity lineups for cross-race culprits and shorter exposures. METHOD: = 2,644) recruited via Prolific watched four short or long videos (half with White, half with African American culprits), completed a distractor task, and made lineup identifications (culprit-present or culprit-absent) with confidence ratings. RESULTS: Both low- and high-SF similarity lineups produced higher discriminability than MTD-only lineups. While low-SF similarity lineups enhanced guilty suspect detection, high-SF similarity lineups better protected innocent suspects. Furthermore, low-SF similarity lineups yielded the best identification accuracy, and high-SF similarity lineups yielded the best diagnosticity ratio. No interactions emerged. CONCLUSIONS: Using match-to-suspect beyond MTD-especially at the highest or lowest SF similarity-improves lineup performance regardless of factors outside the justice system's control. (PsycInfo Database Record (c) 2026 APA, all rights reserved).